Views
No views yet
catsbench, a standalone package that includes benchmark definitions, evaluation metrics, and reusable utilities, including a Triton-optimized log-sum-exp (LSE) matmul kernel.pip:pip install catsbench1from catsbench import BenchmarkHD
2
3bench = BenchmarkHD.from_pretrained(
4 "gregkseno/catsbench",
5 "hd_d2_s50_gaussian_a0.02_gaussian",
6 init_benchmark=False, # skip heavy initialization at load time
7)1x_start = bench.sample_input(32) # [B=32, D=2]
2x_end = bench.sample_target(32) # [B=32, D=2][!IMPORTANT] This samples independently from the marginals, i.e., $(x_0, x_1) \sim p_0(x_0)p_1(x_1)$.
x_start, x_end = bench.sample_input_target(32) # ([B=32, D=2], [B=32, D=2])1x_start = bench.sample_input(32) # [B=32, D=2]
2x_end = bench.sample(x_start) # [B=32, D=2][!NOTE] See the end-to-end benchmark workflow (initialization, evaluation, metrics, plotting) innotebooks/benchmark_usage.ipynb.
1@inproceedings{
2 carrasco2026entering,
3 title={Entering the Era of Discrete Diffusion Models: A Benchmark for Schr\"odinger Bridges and Entropic Optimal Transport},
4 author={Xavier Aramayo Carrasco and Grigoriy Ksenofontov and Aleksei Leonov and Iaroslav Sergeevich Koshelev and Alexander Korotin},
5 booktitle={The Fourteenth International Conference on Learning Representations},
6 year={2026},
7 url={https://openreview.net/forum?id=XcPDT615Gd}
8}